What is ShapedQL?

ShapedQL is a SQL engine built for relevance — it compiles simple SQL queries into real-time ranking pipelines that retrieve, filter, score, and reorder results based on live user behavior. Instead of gluing together Pinecone, Redis, and Python scripts, you can power "For You" feeds, Search, and RAG memory with just 30 lines of SQL. It replaces thousands of lines of infrastructure with native multi-modal embeddings and automated MLOps, helping you build real-time decisions rather than just document retrieval.

Who it's for

• Search engineers — who need to move beyond keyword matching and build semantic, behavior-aware search without managing separate vector databases.
• Recommendation system builders — who want to replace complex pipelines of collaborative filtering, content-based filtering, and reranking with a single SQL-based workflow.
• AI application developers — who are building RAG systems or personalized feeds and need real-time ranking that adapts to user interactions without manual model retraining.

Key features

SQL-based relevance pipelines

Write familiar SQL to define the entire ranking flow — retrieval, filtering, scoring, and reordering — all in one query. ShapedQL compiles that SQL into a production-ready pipeline that runs in real time, eliminating the need for separate infrastructure components.

Native multi-modal embeddings

The engine supports text embeddings for titles and descriptions, image embeddings for visual content, collaborative embeddings for user-item interactions, and personnel embeddings for cast and crew. You can combine these in a single query to blend semantic, visual, and behavioral signals.

Automated MLOps

ShapedQL handles model deployment, versioning, and monitoring automatically. The demo model includes a click-through rate predictor trained with LightGBM, showing how scoring models integrate directly into the SQL pipeline without manual orchestration.

Live playground with demo dataset

The ShapedQL Playground lets you test queries against the Movielens dataset, enriched with IMDb data. You can select use cases like Agent search, Search and feeds, or Recommendations, pick saved queries or write your own, and see results instantly.

What stands out

"Replace thousands of lines of infra with 30 lines of SQL."

That's the core promise — ShapedQL collapses what typically requires a stack of vector databases, caching layers, and custom scoring scripts into a single SQL query. The playground demonstrates this concretely: you can switch between use cases like search, feeds, and recommendations, each powered by the same engine, using the same SQL syntax. The built-in embeddings for text, images, and collaborative signals mean you don't need to manage separate embedding pipelines or worry about keeping them in sync.

Worth checking out if…

You're tired of maintaining brittle infrastructure that glues together vector search, caching, and ranking logic — and you'd rather express your relevance logic in SQL. If you're building personalized feeds, semantic search, or RAG systems that need to adapt to user behavior in real time, ShapedQL offers a radically simpler approach. The playground is a good place to start: pick a use case, run a query, and see how 30 lines of SQL can replace what used to require a dedicated team and multiple services.